Comparing the two methods

Both techniques have their advantages and disadvantages. With adaptive subtraction, the primaries are assumed to have minimum energy because the filters are estimated in a least-squares senses [equation ()]. In addition, adaptive subtraction is sensitive to modeling errors (Chapter ). The strengths of adaptive subtraction are the computing cost and its ease of use, where the filter and patch sizes are only needed. On the contrary, the pattern-based approach is more costly (three to four times) because three inversions need to be run: one for estimating the noise filters N, one for estimating the signal filters S, and one for estimating the primaries [equation ()]. The choice of parameters can be also cumbersome because filter and patch sizes are needed for the two sets of filters (noise and signal). In addition, a signal model usually estimated with the Spitz approximation is needed. The strengths of the pattern-based approach are its robustness to modeling uncertainties and the fact that it does not assume that the signal has minimum energy. These two qualities are particularly important wherever the multiple model is not accurate enough, as it is the case for the dataset presented in this Chapter.